Center
for InformationSystems
Research
Massachusetts Institute ofTechnology
AlfredP, SloanSchoolof Management
50 Memorial Drive
Cambridge, Massachusetts, 02139
CENTRALIZATION VERSUS DECENTRAJ.IZATION
OF INFORMATION SYSTEMS:
A
CRITICAL SURVEY ANDM
ANNOTATED BIBLIOGRAPHYby
Jacob Akoka
November 1977
CISR #36
Sloan WP #1003-78
updated annotated bibliography
by Christine Bullen
Introduction
The literature dealing with the issues related to distributed
manage-ment information systems is rapidly expanding. This literature may be
classified in two categories:
• the management-oriented literature which is concerned
with the issue of centrali;;ation versus decentralization
of information systems and the management of distributed
processing;
a the technically-oriented literature which deals with issues
such as optimizing distributed networks and currency
con-trol in computer networks.
This survey deals only with the first aspect of distributed systems.
In the first part of this paper, a critical study of the state of
the art in distributed management information systems is provided. In
the second part, an annotated bibliography related to the issue of
cen-tralization versus decentralization of information systems is presented.
This bibliography updates the one published by the Center for Information
Systems Research (CISR) (1976) (1)
.
(1) Rockarc, J.I', and Leventer, J.S., "Centralization versus
Decentraliza-tion of Information Systems: An Annotated Bibliography, CISR Report 22,
April 1976
-2-I. CENTRALIZATION VERSUS DECE^^'RAL_IZATION OF INFORMA.TION SYSTEMS:
A
CRITICAL SURVEYThe issue of centralization versus decentralization of computer
re-sources is not a new one; it has been widely discussed and hotly debated for
at least two decades now. The interest in this issue was originally motivated by
the feeling that the computer, a costly expense in terr.s of investment and
operat-ing budget, should be used to the fullest possible potential. Interest also grew
because it was felt that within a corporation, a large measure of political
power rested with whomever controlled the data processing facility. Lately
advances in network technology and the advent of efficient low cost mini and
micro computers has initiated the era of distributed data processing and in
effect thrown new fuel into the centralization/decentralization fire.
Of the voluminous literature published on this subject, we first concentrate
on key articles relating to one aspect of the problem: the
centralization/de-centralization decision. Management, faced with decisions regarding proper
long range directions toward optimal configurations of hardware, software, and
personnel finds little by way of guidelines to follow. There seems then to be
a real need for a rigorous decision model to provide management with an
approach to solving this dilemma.
Ernest Dale (2) states: "the proper balance between centralization and
decentralization often is decided by necessity, intuition, and luck because of
the immense variety of possible human behavior and vast multiplicity of
minute, undiscoverable causes and effects that cannot be encompassed in
(2) Dale, E. '"Centralization versus Decentralization," Advanced Management,
- 3
-any principal or standard of evaluation." In addition, current solutions seem
highly dependent on the characteristics, philosophies, and objectives of
the particular organization for which the decision is to be inade. Accord-ing to George Glaser (3) , "the organizational approach to data p'-ocessing
should be consistent with the overall* organizational approach of the
com-pany in which it functions." The problem is not only of major importance
but of substantial complexity also.
Having surveyed many articles available in the literature it is clear
that, with few exceptions, most articles fit one of the following categories:
• a general discussion of advantages and disadvantages of various configurations as viewed from a decision-making perspective;
• the establishment of decision criteria from specific
cor-porate functions;
• a proposed decision model by which management can m.ake
qualitative decisions about organizational directions
based on specific data processing applications;
« a discussion of distributed systems as being a new and
attractive approach to the
centralization/decentraliza-tion decision.
The first groun of articles is very general and focusses on discussions
of advantages and disadvantages of various ccif igurations. From a
func-tional point of view, cost applications could be accomplished by either
(3) Glaser, G., "The Centralization versus Decericralization Issue: Ar^^'inants
,
- A
-centralized or decentralized approaches. However, as G.A. Champ ine (4)
states, "each of the two approaches has advantages and disadvantages.
In general the advantages of a centralized approach are the disadvantages
of a distributed approach and vice versa." For example, some of the
ad-vantages and disadvantages he lists are:
"Centralized advantages/distributed disadvantages"
o Operations economy
• Hardware economy of scale
• Unified control
• Easy interfile communications
• Easy update/retrieval • Compatibility
"Distributed advantages/centralized disadvantages"
• Communication failsoft capability
• Central site failsoft capability
• Lower communication data rate and costs
• Configuration flexibility
• High speed performance (fast response and high
transaction rate)
• Modular upgrade
Dozens of authors have written similar articles citing specific adva-" j,-s
and disadvantages. Some of these articles ar^i described below:
Rejmolds (5), argues that three economic considerations have to
be taken into account: personnel to operate the hardware, data processing
(4) Champine, G.A., "Six Appraoches to Distributed Databases," Datamation,
May 1977
-5-applications progranming efforts and the conputing. All three considerations can lead to some economic saving when implemented in a centralized way.
Reynolds uses his own organization, Hughes Aircraft Corporation, as an
illustration of his argument.
Kieder (6) argues that two considerations are critical in arriving
at the most effective type of organization for a particular corporation -i.e. corporate structure itself (irrespective of data processing tasks
performed) and size and location of tKe corporation.
Wofsey's (7) article is mainly a discussion of the respective
ad-vantages and disadvantages of both the centralized and decentralized approaches to systems design.
Finally, Burnet and Nolan (8) argue that the technology has now
matured to the stage where the cost of using a mini for certain data
processing jobs compares favorably with using a portion of tlie capacity
of a large machine.
In some articles this approach takes a more general form. Louis
Fried (9) exemplifies this in his article when he states, "As part of
the continuing discussion that is almost as old as the computer industry,
there have,been as many reasons advanced for decentralization as for
centralization. However, in contrast to the arguments for centralization, which center around efficiency, the arguments for decentralization center
around effectiveness." It is my contention that this first group of articles ±s too geieral and diverse from which to draw -any meaningful generalizations
(6) Kieder, S.P., "Once again Centralize or Decentralize," Infosystems,
December 1976
(7) Wofsey, M.M. , "Centralization versus Decentralization," Manage-ent
of EDP Syste~.3 , 1973
(8) Burr.et, G.J., and Nolan, R.L., "At Last Major Roles for Minicomputers,"
Harvard Business Review, May-June 19/5
(9) Fried, L., "Centralization: To Be or Not To Be," Infosystems. January
- 6
-in terms of deoleions regarding optn.^1 solution. As Rockart -•
et. al. (10) state, "The articles on the advantages and disadvantages
of centralization and/or decentralization abound in the literature.
Since different authors have different assumptions and approach the
prob-lem somewhat differently, their arguments are not strictly comparable."
The second group of articles approaches the discussion of centralization
versus decentralization in terms of corporate functions. These articles
are far more useful in that they propose specific ways of looking at the
decision. Norton (11) reiterates this point by stating,
"Generaliza-tion is meaningless when applied as a generality to information systems.
Indeed, the concept of centralization must be approached in terms of
specific functions which make up operations and management of an
organiza--iion's information.system." Accordingly. Norton groups information systems
related activities into three categories: systems development, systems
operations, and systems management. Each of these categories can be
defined functionally as follows:
Systems Development: This includes system design, the development
of detailed specifications and programs, implementation plans, and maintenance
plans.
Systems Operations; This includes the editing and control of input
and output, updating data files, processing, and the reporting of results.
Systems Management; This includes planning long range directions
and projects, and maintaining control over the entire facility.
(10) Rockart, J.F., and Leventer, J.S., "Centralization versus
Decentral-ization of Information Systems: A Critical Survey of Current Literature," CISR Retport 23, April 1976
(11) Norton, D.?.,
"Inromaticn
Syster.s Centralization: The Issues,"
-7-He then goes on to more rigorously define these activities and observes
that the administrative planning and control tasks undoubtedly have more influence on the effectiveness and efficiency of an information system
than other variables. Carl H. Reynolds (12) takes a similar approach
to that of Norton's. Ke divides data processing facilities into three
com-ponents: "the computing hardware," "personnel required to operate the
hard-ware," and "data processing applications programming efforts." These
categories less rigorously define the activities of a data processing facility
and are therefore less useful.
This approach, by which the problem is divided into smaller pieces,
leads to the third category which consists of only one article. Rockart
et. al. (13) follow Norton's reasoning that activities performed by information
systems are three distinct processes: systems operation, systems management,
and systems development. Since each is an independent process, the decision to centralize or decentralize can be made independently for each one. The
authors further segment the problem by looking at the decision in light of the
applications being performed. Their proposal is then basically that decisions to centralize or decentralize can be made separately for each of Norton's
processes (system development, system operations, and system management)
and each group of closely related applications of being performed. Rockart's
model does offer general guidelines for management to follow. It takes a
step in the right direction in that the model proposes concrete procedures
to follow. Although Rockart relies on mainly qualitative methods of
evaluation, his division of decisions with regard to applications opens
the door to quantitative evaluation methods. .
(12) Reynolds, C.H., op. cit.
(13) Rockart, J.F., Leventer, J.S., and Sullen, C.V., "Centralization
versus Decentralization of Information Systems: A Primary Model
- 8
-The last group of articles discusses distributed data processing as
a new and promising trend in data processing configuration, which could elimi-nate the whole centralization/decentralization problem, John Lusa (14) states,
"Some people are still arguing the comparative merits of centralizing or
decentralizing infosystems activities. While the discussion goes on at a
somewhat academic level, a relatively new phrase, if not necessarily
repre-senting a new concept, may keep the discussion at that level. Distributed
processing has blossomed into major prominence as a technique for
increas-ing the efficiency of a data processing operation to the benefit of the
users." This new trend brought on by network technology and the advent
of low-cost mini and micro computers has indeed created an appealing
al-ternative for certain situations. Other authors such as John W. Luke (15),
Richard G. Canning (16) and Tien Chi Chen (17) , to name a few, take similar
positions in favor of distributed data processing. Distributed processing
(14) Lusa, J. M. , "Distributed Computing Alive and Well", Infosystems,
November 1976.
(15) Luke, J. W., "Unravelling the Confusion of Distributed DP",
Info-systems, December 1976.
(16) Canning, R. G. , "Is Your Future Distributed Systems?", EDP Analyzer,
Vol. 11, No. 8, August 1973.
(17) Chen, C. T., "Distributed Intelligence for User Oriented Compull..o",
_ Q _
may well be the computing phenomenon of the 1980's, and it may well be
the solution to some problems. However, aini computers and network
technology will not solve all data processing problems. And the
funda-mental issue of which data and which coEputer processing should take
place at the outboard end of the systea (decentralized) and which
in-board (centralized) still remains.
In conclusion 1 would like to offer Robert L. Patrick's (18)
obser-vation, "A mini is a good solution, sometimes. Decentralization
—
ordistributed processing, or distributed computing, or whatever
—
is agood solution sometimes, but they are only good solutions to some problems,
As in most things we do, the important work is in deciding whether the
solutions we like fit the problems we have."
In surveying the literature, little was found in the way of hard conclusions. The centralization/decentralization decision process is
still very subjective at best. But the factors involved in making
cen-tralization/decentralization decisions are now coming more clearly into
focus.
(18) Patrick, R.L., "Centralizing Hardware and Dispersing Responsibility",
- 10
-II. CENTRALIZATION VERSUS DECENTRALIZATION OF INFORMATION SYSTEMS:
AN ANNOTATED BIBLIOGRAPHY
Becker,
Hal B.
"Preparing for Distributed Data Processing," DataComm User, January 1977, pp. 22-24
This article focuses on new network planning
phil-osophy. It begins by defining the inforcation
net-work environment, stating the user's requirements,
and then using a design sequence to identify
speci-fic hardware and software products. It's a
particu-larly useful technique for distributed processing
networks.
The author divides user requirements into three
basic categories:
1. information processing, or manipulation of Information in a way that produces the desired result;
2. netv7ork processing, or the movement of
information between the various network
nodes;
3. database or storage of information within
the network in forms convenient to network
users.
Becker,
Hal, B.
The author claims that the users requirements should
be essentially independent of specific hardware,
soft-ware and applications. These functional requirements
are defined in terms of: topology, anticipated volume
of information flow, existing information processing
facilities, response time, availability, and security.
Using the requirement statement as input, the analysis
and design sequence employs an iterative, interacti""
series of steps to derive a workable, optimum,
econom-ically justifiable network configuration.
"Six Steps to Network Analysis and Design," DataComm
User. March 1977, pp. 30-31
In this article, the author arguos that in the
str-Jt-tured approach to network inpiemsntation, tj.e analysis
£nd design sequence involves an iterative, interactive
series of steps. It is iterative in that some or all
- 11 -Blair, James, F. Champlne, G.A. Donaldson, Hamish,
that it consists of a combination of automated algorithms
and manually derived design and decision making processes. The sequence consists of six basic steps: the
require-ment staterequire-ment, physical and logical design, performance
prediction, installation/operation, performance validation
and evolution.
"Distributed Inforiration Systems," Telecommunications,
January 1974, pp. 39-40
Distributed ii'.ronnatlon systems are defined here as a hierarchy of processors linked together, each of which may have human as well as database interfaces. A table
defines the typical characteristics of six types of
dis-tributed information systems, however, the distinction
among the types may blur in actual use. Two examples of
DIS are presented: a communication network for the Texas
Department of Public Safety, and one for Bancsystems Association
of Cleveland, a credit authorization organization. By
utilizing the concept of a hierarchy of processors for communications efficiency, many corporations and
institu-tions are achieving cost effective systens. This will
have an overwhelming impact upon the growing data
com-munications industry.
"Six Approaches to Distributed Data Base Systems,"
Datamation, May 1977
In this paper, it is argued that distributed databases
offer a solution to the very real problems of
geograph-ically distributed organizations which need to preserve
a unified information sharing and processing system.
Computer industry trends, such as falling costs of
pro-cessing and storage compared to more stable
coirjiiunica-tion costs, are responsible for the increasing appeal of
distributed database systems. The paper describes the two basic approaches to distributed databases which are
to replicate the database at each node, which will
work
best where file sizes are small andcommun-ication costs high; or to partition- the database, which
works best when most transactions are to "local" large
files. It is shown that when most transactions are not
to local files and file size is large, centralization of
some form may still be the economically justified answer.
"A Case for Distributed Processing," Data Exchange,
July-August ,1975, pp. 43-45
It is argued here that decentralized management needs
- 12
-have different needs, b) it requries a very strong
manager to impose the sane system en two different
divisions, c) a changing business strategy requires
flexibility. The author suggest one should develop
common systems only where there is a very clear common
need.
Edited
Regarding management, however, what seems to be
re-quired is, according to the author, a sort of dual
management: business management from the user
manage-ment and technical supervision from management services
(i.e. DP). It is claimed that in small to medium sized
companies central control is essential. In larger
organizations more flexibility is possible and the advantages of centralized control must be weighed against potential dangers of over-bureaucratic
inter-ference. In summary, central control of systems
development is usually desirable.
"Distributed Data Systems," EDP Analyzer, Vol. 14,
June 1976, pp. 1-13
In this paper, it is claimed that one of the problems
with distributed systems is deciding what functions
should be distributed, where they should be distributed,
how to control the operation of the hierarchy of
func-tions, and how to insure database integrity. Some
mo-tivations for distributed systems are: 1) to enhance
reliability, 2) to take advantage of falling processing
costs, 3) to handle growing transaction volume, 4) to
perform input validation on the spot. In this paper,
it is argued that there are three main ways of
partition-ing systems:
1. Partitioning Of An Application System. Here the
system designer must search for the natural clus-tering of activities.
2. Partitioning, By Functional Area. Each department
would probably have its Cati minicomputer system,
and the several departmental systems might be tied
together in some sort of network system.
Edited
3. Partitioning Of The Data Processing Function.
Sup-porting remote data entry, handling data
communica-tions functions, or handling the database management.
"Distributed Systems and the End User," EDP Analyzer,
Vol. 14, No. 10, October 1976
The article focuses on the numerous reasons which account
- 13
-Edited
example, one reason is the reduction in overall
op-erating costs that these systens promise. Another
reason is that they give each organizational unit the
resources to do its own data processing - to "control
its own destiny" in the words of one executive. At
the same time distributed systems may well bring
radi-cal change in the whole data processing environment,
as we know it today. Almost all aspects of the data
processing function - systems development, programming,
data entry, and computer operations - may shift in the
direction of the end user. This article includes the experiences of some organizations where such a shift
has already started.
"Distributed Computing and the Mini-Computer," Canadian Data Systems , July 1974, pp. 38-39
This article claims that distributed computing means
putting the computer power where the job is. The
author asserts that by using the right small computer
at the right place, there are increases of control and
a reduction of communications costs. It is argued
that, only summary data needs to be communicated to other
elements in an overall system. This article further
lists areas around a plant where computer power should
be distributed.
Edited "Distributed Minis Score Over One Large System at Equitable," DataComm User, March 1975, pp. 51-52
In this paper, it is reported that the Equitable Life
Insurance Society has implemented a series of
distribut-ed mini computer systems in aspects of its claims
pro-cessing, and statistical analysis. It is argued that
the mini's shorter implementation period, low
implemen-tation cost compared to large batch/on line computer
systems, lower operating costs, and lower site
prepar-ation costs are factors which lead to the decision.
Equitable uses the mini's for processing in real-time
mode for the following applications: group claim payment
r.ys-tem; calculations for group insurance compensations for
agents; administration of medicare for the states of
Idaho, New Mexico, Tennessee, and W^/oming; integrnted
group insurance system, to miintain a unified dat^:! base
with exhaustive premium and claim information for all
policies; and a satellite programming office in Cresskill,
- 14
-Edited : "Talking About Distributed Processing," Data Systems
,
February 1976, pp. 8-23
This article presents a number of different views on
both the principle and practice of distributed
pro-cessing. At the present time, distributed processing
systems are evolving rather than being planned. It
is argued that dispersed unconnected data processing
systems are turned into distributed systems by the
in-stallation of ccmmunication links. Such systems have
evolved throui^L the need to retain vital data
locally, and to improve the operation and control of.
remote systems. Problems connected with the concept
are largely management oriented. The editor stresses
the fact that the planning phase is vitally important
and must be exhaustive and informed. Management must
also solve the problem of security and lack of
uniform-ity of accounting techniques.
Edited : "Talking About Networks," Data Systems, March 1977,
pp. 7-9
Thi§ article indicates that there are now more than
twelve major communication networks, private and
commercial in the world. It is claimed that the implementation of a network raises many problems
be-yond those of its design. The article describes below
some of these problems:
1. Standardization And S:.E:3le Interfaces To All Kinds Of Hardware And To Other :.'etvorks. The
Interna-tional Standards Organization has been largely
con-cerned with this and was responsible for high level
data link control. The issue has been complicated
by IBM's introduction of systems network
archi-tecture.
2. Tariffs. Agreement must be reached between
coun-tries involved in networks which cross national
boundaries. The difficulties' of privacy and data
security increase costs.
3. Demands Of Individual Government Authorities,
Particularly In The Case Of Security And P-^ivacy
On International Networks. The author describes
- 15
-Edited : "Distributed Processing Is In, V;hatever It Is," Datamation,
Vol. 22, December 1976, pp. 102-111
The article recognizes the fact that distributed
proces-sing appears tobe flourishing in Europe. Most plans will
be implemented In the next 3-4 years. Most are planning
to put part of their databases and processing capability
into intelligent terminal systems, microconputers, and
small business systems located at user departments
or remote factory floors, distribution centers, and sales
offices. Reasons tended to center on better services
to end users, followed by cost implications.
It is claimed that the central site generally remains in
ultimate control. Hardware selection and systems program-ming will be done by a central staff. Standardization is
the reason for centralized control. Applications include
order processing and inventor;/ or stock control,
produc-tion control and planning, and local management and
fin-ancial applications. Most remote processors will have a
portion of the database and will be in daily communication
with the host processor.
Emery, : "Managerial and Economic Issues in Distributed
James C. Computing," IFIP Proceedings. 1977, pp. 945-55.
"Discusses technological developments which have„jnade
it feasible to distribute processing functions,
alter-native computing configurations, and the centralization versus decentralization issue. The advantages and
hazards of distributed systems are also examined."
[from Quarterly Bibliography of Computers and Data
Processing, Vol.7, no. 4, January 1978.]
Fledelman, ; "It's A Small World," Infosystems, Vol. 24, April 1977,
Lawrence pp. 50-54
In this article, the author argues that the distributed
system using minicomputers is not the only means to meet
the needs for multi-locaticn data processing; in fact,
it Is in direct compt>.j.i.j.ci with large computer systems
with on-line remote terminrls. However, as the price of
minicomputers continues to drop, extensive business soft-ware enhancements are introduced, data communications
facilities become more economical> «nd the data processing
manager develops more knowledge of how to handle this
sit-uation, we are seeing an ever growing trend toward
dis-tributed crr-utin^. The ccta prtce^sins m-ir.a'.;er vill reed
to determine how such equipment can best be utilized ana
- 16
-Fried, Louis
"Centralization: 1976
To Be or Not To Be," Infcsystems, January
Glaser, George Hannan, James and Fried, Louis Hunter, John J.
This article begins with the description of
Citibank's data processing reorganization toward
decen-tralization. It is claimed that change can be traumatic
and the "average" corporation is advised to avoid it
un-less it is well justified. Arguments are presented citing
advantages and disadvantages on both sides of the
central-ization/decentralization issue. The best structure^in
the end, seems to result from careful analysis of the
total corporate requirements of data
processing. After the many considerations are weighed ,
the best reorganization method should be tailored to fit
an individual corporate situation.
"Distributed Data Processing: Promises and Pitfalls," Proceedings of the IDC Conference on Distributed Data Processing, April 1977
George Glaser begins this article by refering to
dis-tributed data processing as "offering the joys of
computerdom long promised by the industry. As with any
reorganization plan the costs are likely to be high,
but he cites this as similar to the costs of any
tech-nological advance. The author claims that the primary advantage
of
distributed processing is that it wouldim-prove the quality of service rendered by an enterprise,
in that operating needs would be assessed more
real-istically and on a more timely basis. This is not to
say reorganization toward distributed processing is not
without problems. The author asserts that a battery of
issues like spending authorities, development priorities,
quality control, and standards will certainly challenge even the most experienced managers' skills.
"Should You Decentralize," Computer Decisions, Vol. 9,
February 1977, pp. 40-42
This article approaches the question of centralization
versus decentralization from an advantages versus
dis-advantages perspective. Several systems incorporating
the best of both are discussed, and specific criteria
is outlined for making the decision.
"Distributing a Database," Computer Decisions, Vol. 8,
June 1976, pp. 36-40
The author reports that about a decade ago, there was a
switch from remote processing to a new discipline of
cen-tralized commuting. Some problems that came with
cen-tralization were enormous databases, data inavailability
due to line failures, concern for data security,
:_--overloading of the central processor. One answer to these
concerns proposed by D? strategists is distributed
17 -Kelsch, August L. Kleder, Stephen P. La Vole, Paul
points where the transactions occur. This way the user's
data are always available, and there is no worry about
data cocmunications failures.
Possible ways of distributing a database to meet the
de-mands of the three cost coinaonly used distributed
pro-cessing networks - star, hierarchal and ring - are also also discussed.
"Dispersed and Distributed Data Processing," Journal
of Systems Manager.ent, Vol. 29, No. 3, March 1978,
pp. 32-37.
This article focussfes on an organizational design which
"marries" distributed and centralized data processing
In an attempt to "maximize the availability of the
com-puter as a business service tool." The author allocates
operational responsibility to the end user of a
distrib-uted system and retains technical responsibilities in a
centralized DP group.
"Once Again: Centralize or Decentralize", Infosystems,
December 1976.
This article addresses the issue of centralization vs.
decentralization of corporate data processing systems. The article cites two considerations as critical in
arriving at the most effeccive type of organization
for a particular corporation: 1) corporate structure
itself Irrespective of data processing tasks performed; and
2) size and location. The author finishes by citing
several advantages are disadvantages for both systems.
"Distributed Computing, Systematically", Computer
De-cisions. Vol. 9, March 1977, pp. 44-45.
It Is argued that a system approach is mandatory In the
field of distributed data entry and processing. The
author claims that this new field promises important
benefits both to the DP manager and to the user departments.
The author argued that since distributed data entry and processing is a new technique, there are few
stan-dards and even fewer general systems design principles.
This article contains some basic questions
which should help DP management avoid some of these pitfalls.
Luke,
John W,
"Unravelling the Confusion of Distributed Data Process-ing", Infosystems, December 1976.
Tlie author claims that there is considerable divergence
of views on the definition of distributed processing. As a
consequence, it rii-cS the question of vhcthir di-tributei
- 18
-. It is claimed that network services are already
provid-ing distributed processing. Integration of an on-site machine into a network enables utilization of
all the systems resources while retaining control over
the local installation. However, there are still
soft-ware problems to be solved before the full potentialities
of such integration can be realized.
It is argued that commercial networks have a formidable
arsenal of advantages which are offered to their customers.
Re-liability is one of the most valuable features as well as
hundreds of application packages. Thus the author claims
networks offer the best of both centralized and
decentral-ized worlds
.
Lusa, : "Distributed Computing: Alive and Well", Infosystems,
John M. November 1976.
This article suggest that distributed computing is blossoming into a technique of major importance for
increasing the efficiency of data processing
opera-tions. "Distributed Data Processing might be
des-cribed as the marriage of minis and telecoms:unications",
and will mature into the electronic office of the
future.
Patrick, : "Decentralizing Hardware and Dispersing Responsibility",
Robert L. Datamation, May 19 76.
This article states that at one time Grosch's law was more
nearly correct. It may still be useful today when applied to the CPU cluster alone, but support costs,
which are primarily personnel costs, rise more slowly
than the costs of the computer configuration and
there-fore continue the argument that bigger is cheaper.
On the opposing side, the complexity of the entire
sys-tem of hardware, software, and personnel, rises much
more rapidly than computer power or costs. Wnen that
complexity rises to a point where it taxes our abilities
to manage, it argues strongly against centralization
in-spite of any economies of scale.
As a conclusion the author claims: "A mini is a good solu-tion, sometimes. Decentralization or distributed
pro-cessing or distributed computing or whatever -- is a good
solution sometimes, but they are only good solutions to
some problems. As in most things we do, the important
work is in deciding whether the solutions we like fit
- 19
-Re>molds,
Carl H.
"Issues in Centralization", Datamation, March 1977.
This article addresses Hughes Aircraft Company's
de-cision to'centralize their data processing functions.
By doing so it v/as estimated that data processing
costs could be cut by a factor of two.
The article centers on three economic considerations fundamental to making any
centralization/decentraliza-tion decision.
First is the personnel to operate the hardware. Usually centralization offers some savings however as on-line processing comes into greater use, the operator ex-pense will be less an issue. Also the ease with which multiple tasks can be performed by operators at
small-er decentralized installations tends to lessen this expense.
The second issue to be considered is data processing
appli-cations programming efforts. Here one should e>?pect savings
from centralization only if there exists in a
corpora-tion common programming applications. In corporations
with varied application of data processing facilities decentralization can be more economical.
The last issue to be considered is the computing hardware.
Economy of scale does not seem to be as important an
argu-ment for centralizaton as it used to be due to the advent
of inexpensive mini-computers: Tracking down system
failures in smaller decentralized installations is easier
and can in fact mean additional savings in data
proces-sing costs.
Ritchie,
Robert 0.
It appears that the centallzation resulted in major
cost savings for Hughes Aircraft Company in data
processing functions. Some effectiveness might have
been sacrificed but in terms of the individual data
processing needs of the company, centralization was
detejcmined to be the better organizational structure.
"Intelligent Terminals and Distributed Processing",
Computer Decisions, February 1975, pp. 36-42.
In this paper it is argued that intelligent terminals
and distributed processii»g provide processing
capa-bilities to remote locations, freeing the large cen-tral computer of menial processing tasks.
Communica-tion costs decrease, fewer I/O's are needed to the data
base or mainframe, and the user has computational power
unique to his requirements as well as a much more
re-liable system and independence from error-prone
communi-cations f'ciiitias. The Euthor gives su~:;e3tion? to
- 20 -Rockart, John F. and Leventer, Joav S. and Bullen, Christine
"Centralization versus Decentralization of Infornation
Systems: A Preliminary Model for Decision-Making",
CISR Report, Sloan School of Management, MIT, Cambridge,
Massachusetts (forthcoming).
Of the many organizational decisions facing management
today one of the most difficult must certainly be wtiether
to centralize or decentralize their data processing
structure. In this area v/here guidelines are general
at best, Rockart proposes a model which if used
proper-ly assesses each pertinent factor involved in a
corpora-tion and logr.cally determines which functions within the
data processiiig facility should be done locally and
which functions should best be accomplished centrally. Thus the model is based on Norton's observation that the activities performed by information systems are in
fact three distinct processes: systems development,
systems operation, and systems management. Since each
of these is an independent process the decision to
cen-tralize or decentralize can be made independently for
each process. Further segmentation with respect to
logical application groups can be made for both systems
development and systems operations, however systems
management is viewed as being a decision made for the corporation-as a whole.
Rockart's decision model though entirely qualitative
offers precise guidelines for m.anagement to follow.
It is the most advanced model in this area.
Severino,
Elizabeth F.
"Databases and Distributed Processing", Computer Decisions,
Vol. 9, March 1977, pp. 40-42.
It is argued here that the concepts of distributed
pro-cessing are maturing. This growth should play a role in
the
decision to install a distributed processingsys-tem. There are two kinds of processing power,
horizon-tal and hierarchical. In horizontal distribution, all
de-vices co-operate at the same logical level to perform
a set of tasks. Rule and control are democratic. The
processes exchange jobs or transactions so that the total
workload is distributed. In a hierarchical scheme of
distribution, the devices are interconnected to form a
functional hierarchy. The sharing of tasks and other
functional relationships are structured and all devices are
controlled by a primary network computer. There are also hybr:
forms of networks that use both approaches and even
- 21 -Statland, Norman and Winski, Donald
"Distributed Information Systems: Their Effect on Your
Company," Price Waterhouse Review, Vol.23, No. 1, 1978,
pp. 54-63.
This article examines the organizational and management
implications of the distributed processing environrr.ent
.
The authors point out that too often the general manager
has viewed the MIS department as an overhead cost which he/she wants to minimize, rather than seeing it as an
in-vestment to be maximized. The authors propose a distrib-uted processing scenario which plans systems development
and operations with the user and maintains a centralized
MIS department for promulgating standards and procedures.
Woods,
Larry D.
"Distributed Processing in Manufacturing," Datamation,
Vol. 23, October 1977, pp. 60-63.
"Addresses the potential for the loss of control and
the problems of maintaining corporate data bases in
light of the proliferation of minicomputers in a large
manufacturing company. Specifically, the environ,menr.
of the John Deere Company and its approach to solving
those problems are discussed." [from Quarterly
Biblio-graphy of Computers and Data Processing, Vol.7, no.4,
- 22
-LIST OF THE ARTICLES SURVEYED
(1) Becker, Hal B., "Preparing for Distributed Data Processing", Data
Comm User, January 1977, pp. 22-24.
(2) Becker, Hal B., "Six Steps to Network Analysis and Design", Data
Comm User, March 1977, pp. 30-31.
(3) Blair, James F. , "Distributed Information Systems", Telecoinaunications,
January 1974, pp. 39-40.
(4) Champine, G. A., "Six Approaches to Distributed Data Base Systems",
Datamation, May 19 77.
(5) Donaldson, Hamlsh, "A Case for Distributed Processing", Data Exchange,
July-August 1975, pp. 43-45.
(6) Edited, "Distributed Data Systems", EDP Analyzer, Vol. 14, June 1976,
pp. 1-13.
(7) Edited, "Distributed Systems and the End User", EDP Analyzer, Vol. 14,
No. 10, October 1976.
(8) Edited, "Distributed Computing and the Mini Computer", Canadian
Data-systems, July 19 74, pp. 38-39.
(9) Edited, "Distributed Minis Score Over One Large System at Equitable",
Data Comm User, March 1975, pp. 51-52.
(10) Edited, "Talking about Distributed Processing", Data Systems, February
1976, pp. 8-23.
(11) Edited, "Talking About Networks", Data Systems, March 1977, pp. 7-9.
(12) Edited, "Distributed Processing Is In, Whatever It Is", Datamation,
Vol. 22, December 1976, pp. 102-111.
(13) Emery, James
C,
"Managerial and Economic Issues in DistributedComputing," IFIP Proceedings,
i""'-'
pp. 945-55.
(14) Feidelman, Lawrence, "Its a Small World", Infosystems, Vol. 24,
April 1977, pp. 50-54.
(15) Fried, Louis, "Centralization: To Be or Not To Be", Infosystems,
January 1976.
(16) Giassr, George, "Distributed Data Prccass'.-.s: Prc-i^es 2r.d
Pit-falls", Proceedings of the IDC Conference on Distributed
23
-(17) Hannan, Jaiies and Fried, Louis, "Should You Decentralize," Computer
Decisions, Vol, 9, February 19 77, pp. ^0-42.
(18) Hunter, John J., "Distributing a Database", Computer Decisions,
Vol. 8, June 1976, pp. 36-40.
(19) Kelsch, August L., 'Dispersed and Distributed Data Processing," Journal
of Systems Management, Vol. 29, No. 3, March 1978, pp. 32-37.
(20) Kieder, Stephen P., "Once Again: Centralize or Decentralize",
Info-systems, December 1976.
(21) La Voie, Paul, "Distributed Computing, Systematically", Computer
Decisions, Vol. 9, March 1977, pp. 44-45.
(22) Luke, John W. , "Unravelling the Confusion of Distributed Data
Processing", Infosystems, December 1976.
(23) Lusa, John M. , "Distributed Computing: Alive and Well", Infosystems,
November 1976.
(24) Patrick, Robert L., "Decentralizing Hardware and Dispersing
Responsi-bility", Datamation, May 1976.
(25) Reynolds, Carl H., "Issues inCentralization", Datamation, March 1977.
(26) Ritchie, Robert 0., "Intelligent Terminals and Distributed
Process-ing", Computer Decisions, February 1975, pp. 36-42.
(27) Rockart, John F., Leventer, Joav S., and Bullen, Christine V,,
"Cen-tralization versus Decentralization of Information Systems:
A Preliminary Model for Decision-Making", CISR Report, Sloan
School of Management, MIT, Cambridge, Massachusetts
(28) Severino, Elizabeth F, "Database and Distributed Processing", Computer
Decisions, Vol. 9, M^rch 1977, pp. 40-42.
(29) Statland, Norman and Dor.ald Winski, "Distributed Information Systems:
Their Effect on Your Company," Price Waterhcuse Review, Vol. 23,
No. 1, 1978, pp. 54-63.
(30) Woods, Lariy D., "Distributed Processing in Manufacturing," Datamation,